http://wiki.apache.org/cassandra/FAQ#range_ghosts
On Sun, Apr 11, 2010 at 5:12 PM, Kevin Wiggen <kwig...@xythos.com> wrote:
>
> I have spent the last few days playing with Cassandra and I have attempted
> to create a simple "Java->Thrift->Cassandra" Discussion Group Server
> (because the world needs another one) to teach myself the data model and try
> everything out.
> With all the great blog posts on cassandra out there, I am now able to
> read/write/delete/modify a nested discussion server. YEA!!!
> I decided to have two simple ColumnFamilies.
> One called Posts
> Post = {
> '7561a442-24e2-11df-8924-001ff3591711': { //UUID
> 'id': '7561a442-24e2-11df-8924-001ff3591711', //ID ==
> UUID
> 'parent_id': '89da3178-24e2-11df-8924-001ff3591711' //Parent
> Post UUID
> 'author': 'a4a70900-24e1-11df-8924-001ff3591711', //Users
> UUID
> 'subject': 'This is a forum post', //Subject
> 'body': 'Forum post body. This is awesome!', //Body
> '_ts': '89da3178-24e2-11df-8924-001ff3596713', //TimeUUID
> },
> }
> Where the key is a simple UUID and the columns are the Forum/Post/Replies.
> A Forum has a hardcoded Parent UUID which I store in Java, while the Posts
> and Replies are tied to their parent posts/forums/etc by the parent_id. I
> sort by UTF8Type, but it really doesn't matter in this case as I drive into
> this map always by the Key and always get all columns (6 of them).
> All queries drive into the second ColumnFamily called Threads
> Thread = {
> '7561a442-24e2-11df-8924-001ff3591711': { //Parent
> thread UUID
> #timestamp of post: post UUID
> '89da3178-24e2-11df-8924-001ff3596713':
> '7561a442-24e2-11df-8924-001ff3591711',//TimeUUID column name -> post
> UUID value
> },
> }
> With a Parent UUID I can drive into Threads which will give me the list of
> Posts/Replies at that level sorted by TimeUUID. Column name is the post
> TimeUUID and the value is the Post UUID. This ColumnFamily is sorted by
> TimeUUID.
> Thus I can walk the tree (of any depth) of Forum/Post/Replies with the
> Thread table.
> I have this all working on a single cassandra node and it works great.
> Inserts go to both tables while deletes need to use the Thread ColumnFamily
> to recursively delete all child posts, the Column in the Parent key of
> Thread and all associated data in Post.
> Any comments on whether this is a good/terrible data model, etc so far are
> welcome. :)
> My question comes from the fact that during this process I have
> written/read/deleted many "key->Columns" to these ColumnFamilies (many of
> which failed half-way through) so I decided to write a "clean" script to
> remove all data from these ColumnFamilies (much like a truncate table
> command in SQL).
> Using the following Java code
> //get the ID column for each KEY we find
> List<byte[]> l_columns = new ArrayList<byte[]>();
> l_columns.add(Transcoder.encode(ID));
> SlicePredicate l_slicePredicate = new SlicePredicate();
> l_slicePredicate.setColumn_names(l_columns);
> //get 100 keys at a time
> KeyRange keyRange = new KeyRange(100);
> keyRange.setStart_key("");
> keyRange.setEnd_key("");
> List<KeySlice> l_keySlices =
> p_context.getClient().get_range_slices("Discussions", new
> ColumnParent("Posts"),
>
> l_slicePredicate, keyRange, ConsistencyLevel.ONE);
> I get ALL of the KEYS I ever wrote to the server. Most of them have no
> Columns associated with them. In fact if I query the same key with
> SlicePredicate l_slicePredicate = new SlicePredicate();
> SliceRange l_sliceRange = new SliceRange();
> l_sliceRange.setStart(new byte[] {});
> l_sliceRange.setFinish(new byte[] {});
> l_slicePredicate.setSlice_range(l_sliceRange);
> List<ColumnOrSuperColumn> l_result =
> p_context.getClient().get_slice("Discussions", <KEY FROM
> GET_RANGE_SLICES>, new ColumnParent("Posts"),
> l_slicePredicate,
> ConsistencyLevel.ONE);
> it returns a empty array list (the same if I give it a KEY it has never
> seen).
> It is OK with me if get_range_slices returns keys with no columns (although
> it makes it a little harder to explain to others -- is there garbage
> collection that will clean these out in the future?), however I am stuck on
> how to simply truncate the table without looping through all the values
> looking for something that has a Column associated with it and then deleting
> that key->value.
> It is possible I am not deleting correctly as well. For that I simply do:
> p_context.getClient().remove("Discussions", p_postUUID.toString(),
> new ColumnPath("Posts"), l_rightNow,
> ConsistencyLevel.ALL);
> Just trying to understand what I am getting and compare it against what I
> expected. I am also still trying to write a simple "clean" command.
> If you read this far, thanks.... If you can add some clarity it would help
> me. I have tried to find it in archives and blog posts, but I didn't see
> anything.
> Thanks,
> Kevin
>
>
> This email and any attachments may contain confidential and proprietary
> information of Xythos that is for the sole use of the intended recipient. If
> you are not the intended recipient, disclosure, copying, re-distribution or
> other use of any of this information is strictly prohibited. Please
> immediately notify the sender and delete this transmission if you received
> this email in error.
>
